How Twitter Handles Massive Traffic Surges with Stress Testing and Preparedness

Twitter often experiences sudden spikes in system load due to hot events, such as the “Sky City” incident in Japan that pushed tweet volume from about 10,000 per second to 34,000 per second, yet the platform remained stable.

In an InfoQ interview, Twitter’s VP of Engineering highlighted two key practices: rehearsal (pre‑run) and contingency planning.

Stress Testing

Twitter conducts extensive regular stress tests, simulating extreme conditions and analyzing outcomes to build response plans. Monthly they run a full‑scale stress test and weekly they review performance metrics of the whole system and individual services, ensuring they understand current capacity.

They discuss whether the system is operating efficiently, whether the current number of servers can support expected product load, and whether additional machines are needed.

If a service shows abnormal request counts, it is examined closely and adjusted as needed.

When an unprecedented event like “Sky City” occurs, the prior stress testing has already pushed the system to that level, turning the incident into a real‑world validation.

Extreme Testing

Product teams continuously test extreme scenarios, including fault injection that randomly kills machines in a data center while demanding the service remain available.

Formulating Contingency Plans

Through extensive testing, Twitter documents various failure cases and their remedies, creating run‑books that describe each system’s operating conditions and failure triggers.

Although it is impossible to anticipate every scenario, the documented issues and solutions, along with a “glass wall” that records key information, enable rapid decision‑making during incidents.

Preparing in advance and defining response procedures are essential for maintaining stability.